Locally Weighted Adjustable Parameter-Based LPVG in the Identification of Functional Regions

Abstract

Mobile users’ online traffic records are a type of data that are closely related to human activities. To explore the differences in lifestyles of Internet users, online records of base stations expressed as time series need to be classified. However, due to the circadian rhythms, such time series have the same trend and small-scale local differences. Traditional similarity measurement using original time series is difficult to adequately capture local features of time series. By mapping the time series into visibility graph, the similarity measurement in the graph domain is obtained preserving the dynamic structure and local features to the greatest extent. In this paper, a locally weighted adjustable parameter-based limited penetrable visibility graph (LWAPLPVG) is proposed to further improve the noise resistance of visibility graph and local identification of limited penetrable visibility graph. We get APLPVG by modifying the visibility criteria, which has noise resistance while preserving local features as well. Simple measurement, such as degree sequence in graph domain, is extracted and proved, which is able to amplify small-scale differences. By considering the circadian rhythms of cyclical and trend time series, APLPVG is weighted locally. We use a real dataset of usage detail records (UDRs) to verify that LWAPLPVG can better improve the identification of time series of functional regions with noise resistance.